Creative and media companies across the industry are looking to adopt AI and are encountering challenges in how to govern it. Those are some of the key findings that the study Creative UK has just delivered to the Department for Culture, Media and Sport. It draws on focus groups with 20 micro and SME businesses, as well as expert panels of senior technology leaders. The evidence also includes six case studies of advanced adopters such as Framestore. The researchers went in looking at technology adoption. What they found was a hiring gap, and it’s the one we work in every week.
The appetite was never in question. 64% of creative micro-businesses conducted R&D over the past three years (Siepel, 2025). Arts Council England backed 194 AI projects between 2019 and 2025, worth close to £4 million (Murphy et al., 2026). The tools are cheap, and the budgets exist. What keeps stalling is the step from using AI to governing it, and that step is a person most companies haven’t hired.
Key Takeaways
- A new Creative UK study for DCMS found AI adoption across creative and media stalls on leadership and governance, not on tools
- The study’s expert panels called AI a change project rather than a tooling decision, the same line we draw between studios that scale and studios that stall
- A capability divide is opening between companies that can govern AI and those that can’t, and it compounds with every project
- The recommended fix is capability building, including leadership. In hiring terms, buy judgement, not just AI skills
Caution stands in for a decision
The companies in the study aren’t dabbling. They report faster iteration, lighter admin, better cost forecasting and increased clarity on business operations. The report talks about AI clearing “digital drudgery”, so the team gets more time on the work that wins business. One case study put the commercial logic in a sentence. “If creatives have more time to do the original thought, we win more business.”
The gains they describe aren’t speculative. Businesses already using AI report that it is doing the heavy lifting on often repetitive work. That reduces the admin burden, first drafts, and version control. This enables teams to focus on more productive work that ultimately generates business revenue. That’s the whole appeal in creative and media, where the margin sits in original ideation rather than volume output. Yet the same businesses stop short of putting AI anywhere near the finished product. They have felt the upside and drawn a clear line, but it is drawn out of caution rather than strategy, because no one has worked out where it should enter the workflow.
The rollout runs the same way everywhere. Back office first, client work later, and only once the company can stand behind the output. Creative businesses sell originality and trust, so one low-quality AI-related project risks jeopardising more than their reputation. These companies aren’t slow because they’re nervous. They’re slow because nobody in the building can tell them when it’s safe to move faster.
The tools were never the obstacle
That phrase comes from the study’s expert panels, CTOs and innovation directors at larger organisations, and it’s the most useful line in the document. Most creative and media SMEs can’t validate a vendor claim, can’t see how a tool reshapes a workflow, and can’t price the risk to IP, data or client trust. So, AI gets trialled outside any formal technology function, with no evaluation and no governance, and it stays a side experiment.
The risk sits one step further in. Unvalidated AI output often moves the work sideways rather than reducing it. The team moves from producing to verifying. A junior generates a first pass in seconds. A senior then spends an hour confirming it hasn’t invented a fact, breached a licence or drifted off brief. The savings the tool promised would go back into policing what it made. The study found that this is where adoption quietly stalls. It rarely shows up as a visible failure. It comes through the slow recognition that output nobody trusts costs as much to check as it saved to produce.
A clear position means more than a yes or a no. It means someone senior saying where AI is welcome, where it is not, and what has to be true before that answer changes. Without it, every individual makes the call alone, and the safest solo call is always to do nothing.
We made the same argument about what the best AI talent sees, where 95% of enterprise AI pilots failed on structure and hiring rather than model quality. The Creative UK work puts UK evidence under it. The tools were never the bottleneck. The judgement around them is.
No one wants first, but no one wants last
The study’s sharpest warning concerns divergence. Companies that build governance and have somewhere safe to experiment integrate faster and at a greater scale. The ones that don’t pay more and risk more every time they try, and the gap widens with each project (Creative UK for DCMS, 2026). Each governed project teaches the business something the next one uses, so the lead grows from experience, and the slow to adopt have no way to shortcut.
The divide is widening even where companies think they have opted out. The software they already license is building AI into its core functions, so the easiest route through a workflow becomes the default one, whether or not anyone decided it should. A studio that has taken no position is still adopting AI, just without the governance to direct it. Adoption by default puts a company on the wrong side of the divide, even as it believes it is sitting out the question.
A rival can buy the same tools tomorrow. It can’t buy the year of structured judgement, research, and development that makes those tools generate ROI, because that advantage is organisational, not technical. One case study called the cost of standing still an “opportunity cost so large it’s existential”. On this evidence, that’s arithmetic, not drama.
Framestore made governance the product
Framestore, the most advanced adopter in the study, runs a formal approval process for AI and teaches its clients the difference between workflow optimisation and generative output. That’s governance doing commercial work. The same client demands AI speed on one project and bans it from the finished frame on the next. The position is written into the contract. A studio holds both at once and defends each account without losing either.
The details within that process are what most companies have yet to work through. It covers which licences permit commercial use. It flags which models present the highest risk because their training data can’t be accounted for. And it sets where staff can experiment without exposing client material to a tool that might retain it. The line Framestore draws between improving how the work is made and generating the work itself is where the commercial risk concentrates. Governance of this kind is a sequence of decisions a senior person makes and stands behind. A policy document filed and forgotten does none of that work.
None of this rests on owning better AI. Framestore’s advantage is the process wrapped around ordinary tools, and a process is far harder to copy than a licence.
Aardman, another of the study’s advanced adopters, directs the tools at the work its craft doesn’t need to carry and keeps its people on the part clients pay for. The instinct is the same one Framestore codified, reached by a different studio in a different corner of the market. Nobody expects a small studio to copy a VFX giant. The point is the judgement, not the headcount, and in most companies it isn’t in the building.
Critical thinking beats tool fluency
The study’s recommendation is capability building, including leadership, not just training the workforce. We’d go further. Generic business support and innovation funding aren’t landing, on the report’s own account, because creative businesses don’t see them as built for project-based, IP-led work. What closes the gap is a hire.
And rarely the obvious one. In the briefs we see, the spec usually asks for the wrong thing. A tools person, a prompt wizard, a head of innovation with a showreel. The hire that works looks duller on paper and runs deeper. Someone who has delivered client work under real deadlines and thinks critically about where AI helps and where it quietly hurts. They can work out how it lands in this business rather than in a vendor demo. The production years matter. The rarer asset is the critical thinking to question an output and the judgment to know how to put AI to work.
Hiring junior-level AI skills doesn’t solve the problem either. A capable prompt engineer, two years out of a degree, can run the tools. They still have no standing to tell a client what the studio will and won’t do with their footage. Governance is a question of seniority before it is one of skill.
The wrong hire is expensive, in a quiet way. A pure technologist arrives fluent in models and blank on how a job moves from brief to delivery. They sharpen the parts that were never the problem and leave the client-facing risk untouched. The studio gets faster at the wrong things and no safer at the ones that lose accounts. The person who works has sat in the room when a deadline slipped. They have watched a client change the brief at the last minute. They know which corners AI can take and which it cannot. That judgement is not on a CV or resume under a heading. It’s the thing the brief should ask for and almost never does.
The next objection is cost. A studio of twelve can’t support a full-time technology director, and the report is blunt: most can’t. That misreads what the work needs, senior judgement applied to a dozen decisions, not a permanent seat and a six-figure salary. Some of it arrives fractional, some interim, some folded into a commercial hire who already has the production years and takes the AI brief on top. Set against an opportunity cost, the report’s own case studies call for an existential judgement; the judgement costs less than the alternative.
Nobody advertises a Head of AI
Most of these hires happen quietly, for the reasons we set out in how creative tech is discreetly building its AI capability. The companies moving fastest aren’t announcing a Head of AI. They’re bringing in a leader who can put governance in place, set the evaluation bar, and decide where AI belongs in the pipeline and where it doesn’t.
The quiet is deliberate. A public Head of AI search tells competitors and clients exactly where a company thinks it is behind, and in a market that sells confidence, that is a cost in itself. The businesses getting it right treat the appointment as they would any senior commercial hire, making it early and announcing it to no one until the work is already paying.
The companies on the right side of the divide treated AI as a leadership decision and hired as if they meant it.
